madVR - high quality video renderer (GPU assisted)

[hoborg]

Why would you ask the video renderer to do this? Its not meant to control playback.

Request these features from the developers of your player.

Hi.
Because MadVR already have shortcuts and streams switching support (from MadVR icon).
In my point of view, it is easier to support it correctly on MadVR level instead of many media players.

[COOLak]

Have you tried setting image upscaling to bilinear? That will free up resources that you can use for putting neurons in image doubling and chroma upscaling.

Tried it, and all stays the same. If I set more than 32 neurons in any of the NNEDI3 related settings, the framarate becomes unwatchable. The only image upscaling method that works without lags even with all neurons set to 256 is DXVA2, but it results to horrible image quality with lots of aliasing and other artifacts.

Nope, it was the native video artifacts that are present with any settings...
Hmm, so if the image upscaling method is DXVA2, then nothing lags for me, even 256 neurons with quadrupling. But does it even work with DXVA2 upscaling method? Or it only doens't lag because NNEDI3 doesn't even support this?

[Arm3nian]

disable reclock in this case.
when SM is active there is no repeated or dropped frame. it is still shown but it is not going to drop or repeat a frame.

When bitstreaming to my receiver reclock is disabled by default so no problems there. But I was using reclock to get wasapi output when watching through headphones, so I was wondering if there was any way to completely disable what reclock does and just use it for wasapi output. Not to take this too far from the madvr thread, but is there any other way to output wasapi from mpchc? The mpchc audio renderer can do it but it results in choppy audio.

Tried it, and all stays the same. If I set more than 32 neurons in any of the NNEDI3 related settings, the framarate becomes unwatchable. The only image upscaling method that works without lags even with all neurons set to 256 is DXVA2, but it results to horrible image quality with lots of aliasing and other artifacts.

Nope, it was the native video artifacts that are present with any settings...
Hmm, so if the image upscaling method is DXVA2, then nothing lags for me, even 256 neurons with quadrupling. But does it even work with DXVA2 upscaling method? Or it only doens't lag because NNEDI3 doesn't even support this?

What resolution is your screen and what type of content are you watching? If watching 1080p videos on a 1080p screen then image doubling doesn't do anything. Image upscaling shouldn't be very demanding for a 590 with chroma upscaling at 32 neurons (watching in windowed mode for example uses image upscaling). My 750ti can do this. Image doubling on the other hand is very demanding. My watercooled 780ti could do 128 neurons on only chroma upscaling, but you're effectively running a single 580 since madvr doesn't work with sli. I think you should be able to do 64 on chroma upscaling with every other setting like dither, smooth motion etc... but Image doubling is demanding as I said, destroys my 750ti. Also make sure you're doing the tests in fullscreen exclusive mode.

[ryrynz]

I think you should be able to do 64 on chroma upscaling with every other setting like dither, smooth motion etc... but Image doubling is demanding as I said, destroys my 750ti. Also make sure you're doing the tests in fullscreen exclusive mode.

I wouldn't use anything over 16 on Chroma on most hardware, I did some IQ tests and there's no appreciable differences.
Jinc 3 AR is an excellent alternative but NNEDI 16 on Chroma was only slightly more expensive and I just like the sharpness it provides. Going higher than that you're wasting resources IMO, you won't notice a difference.
I'm running a 750Ti also and I'm getting the most out of it with my settings, some people are just blindly setting options without even checking the results..

[huhn]

When bitstreaming to my receiver reclock is disabled by default so no problems there. But I was using reclock to get wasapi output when watching through headphones, so I was wondering if there was any way to completely disable what reclock does and just use it for wasapi output. Not to take this too far from the madvr thread, but is there any other way to output wasapi from mpchc? The mpchc audio renderer can do it but it results in choppy audio.
use MPC BE for this. the WASAPI in MPC BE works great i don't know why MPC HC didn't merge that into MPC HC but this is the wrong thread for this.

[Arm3nian]

I wouldn't use anything over 16 on Chroma on most hardware, I did some IQ tests and there's no appreciable differences.
Jinc 3 AR is an excellent alternative but NNEDI 16 on Chroma was only slightly more expensive and I just like the sharpness it provides. Going higher than that you're wasting resources IMO, you won't notice a difference.
I'm running a 750Ti also and I'm getting the most out of it with my settings, some people are just blindly setting options without even checking the results..

I do agree NNEDI3 looks a bit sharper than Jinc, but I use 32 just because my card can run it plus all the other settings I need for bluray. I haven't done a huge amount of testing but why wouldn't I use 32 over 16? The 750ti is the most power efficient gpu, 16 vs 32 might draw a couple of watts more, and I'll take that for some sharpness. On the flip side, my 780ti is much more power hungry. I could set benching clocks and run 256 neurons, but that would just pull 400 watts and make my room a convection oven. Anyway, not saying anyone actually NEEDS to run higher than 16, but it seemed the other user was having performance issues running higher than 32, and if he payed $ for his gpu he should be able to use it to it's max potential (minus the sli part :p)

use MPC BE for this. the WASAPI in MPC BE works great i don't know why MPC HC didn't merge that into MPC HC but this is the wrong thread for this.

Works like a charm, thanks man!

[Anime Viewer]

Tried it, and all stays the same. If I set more than 32 neurons in any of the NNEDI3 related settings, the framarate becomes unwatchable. The only image upscaling method that works without lags even with all neurons set to 256 is DXVA2, but it results to horrible image quality with lots of aliasing and other artifacts.

Nope, it was the native video artifacts that are present with any settings...
Hmm, so if the image upscaling method is DXVA2, then nothing lags for me, even 256 neurons with quadrupling. But does it even work with DXVA2 upscaling method? Or it only doens't lag because NNEDI3 doesn't even support this?

Its not frames rates you should be looking at - what you want to look at is the reading under render times. The lower the render times the better.

Selecting DXVA2 disables software scaling/rendering, so even if you have NNEDI3 selected it will not be used which is why you see no difference between 16 and 256 neruons.

I believe someone asked you earlier, but I don't recall seeing your reply: Do you have chroma double (and/or chroma quadruple) selected? If you do unselect it. Madshi and others have determined in doesn't provide any significant benefit, and has a huge cost to it. The only one that will provide any noticeable benefit (especially at the resolution of your screen) is the luma double option.


I wouldn't use anything over 16 on Chroma on most hardware, I did some IQ tests and there's no appreciable differences.
Jinc 3 AR is an excellent alternative but NNEDI 16 on Chroma was only slightly more expensive and I just like the sharpness it provides. Going higher than that you're wasting resources IMO, you won't notice a difference.
I'm running a 750Ti also and I'm getting the most out of it with my settings, some people are just blindly setting options without even checking the results..

I do agree NNEDI3 looks a bit sharper than Jinc, but I use 32 just because my card can run it plus all the other settings I need for bluray. I haven't done a huge amount of testing but why wouldn't I use 32 over 16? Anyway, not saying anyone actually NEEDS to run higher than 16, but it seemed the other user was having performance issues running higher than 32, and if he payed $ for his gpu he should be able to use it to it's max potential (minus the sli part :p)

Works like a charm, thanks man!



I completely agree with ryrynz. Like ryrynz I did quite a bit of testing, and I couldn't find any significant improvement in image quality from going above 16 neurons for anything. Also like ryrynz I believe its a waste of resources to set neurons to 32 or higher. Those resources could be put to use for some significantly noticeable image enhancement or feature, or even not used at all and get longer life out of your GPU and other components by not taxing the system as much (which over time could wear the components and warp the nearby thermal compounds. )
Running at a higher setting just because you can doesn't validate doing so. You make reference to needing some settings for blu-ray, but what leads you to believe that is a necessity? Generally all of the power hungry enhancements in madVR are engineered to improve lower quality source files. They often do little to nothing for higher quality sources. You ask why you wouldn't use 32 over 16, and as has been previously stated the answer is because you don't need to, and because there isn't a significant enough (if any) improvement to warrant the higher setting. Just because someone pays a bunch of money for something it doesn't mean he/she is going to get his/her money worth out of it. Just because you can press something to its limits doesn't mean its a good idea to do it.

[Arm3nian]

I completely agree with ryrynz. Like ryrynz I did quite a bit of testing, and I couldn't find any significant improvement in image quality from going above 16 neurons for anything. Also like ryrynz I believe its a waste of resources to set neurons to 32 or higher. Those resources could be put to use for some significantly noticeable image enhancement or feature, or even not used at all and get longer life out of your GPU and other components by not taxing the system as much (which over time could wear the components and warp the nearby thermal compounds. )
Running at a higher setting just because you can doesn't validate doing so. You make reference to needing some settings for blu-ray, but what leads you to believe that is a necessity? Generally all of the power hungry enhancements in madVR are engineered to improve lower quality source files. They often do little to nothing for higher quality sources. You ask why you wouldn't use 32 over 16, and as has been previously stated the answer is because you don't need to, and because there isn't a significant enough (if any) improvement to warrant the higher setting. Just because someone pays a bunch of money for something it doesn't mean he/she is going to get his/her money worth out of it. Just because you can press something to its limits doesn't mean its a good idea to do it.

This makes no sense. According to this, there is no reason to use madvr because you don't need it. As long as I can see the video I'm good to go. Let me watch my blurays on a cracked tv screen set to 360p, I don't need anything more as long as I can tell what's mostly going on.

Wasting resources? So not using my full gpu power isn't a waste but using the potential is... No gpu is going to degrade as long as it is being fed a safe voltage and is properly cooled. Electronic components are meant to run 24/7, running 16 vs 32 isn't going to do anything to your gpu life span. You mention the resources being used for something else, but there isn't anything else for blurays, I already have all of the image enhancing settings turned on/configured.

And as a final note, I can see a little sharpness difference between 16 and 32. Not sure what content or what type of screen you have, but just because you can't see a difference doesn't mean others can't. There is obviously diminishing returns as you go up in neurons, but this exists for everything. If I upgrade from a $500 pair of headphones to a $1000 pair am I going to get 2x better sound quality? Probably not. Doesn't mean there isn't a difference. Just because 32 uses more gpu power for a smaller increase in IQ doesn't mean it's wrong to use it, it's called being an Enthusiast.

Just to add, I notice from your username that you must watch a lot of anime. Animation by nature is already very sharp. Bluray is shot on film, which is naturally grainy. Sharpness goes a long way for bluray from the movies that I have watched. Everyone's needs and wants are different. Saying 32 neurons is useless compared to 16 is ignorant. You're on a laptop, going from 16 to 32 would just create more heat which will burn your legs or table and increase fan noise and drain your battery. Those two effects are not worth a slight increase in an already sharp piece of content. I'm on the most power efficient htpc I could build that runs cool by default, and I am willing to trade a little power draw for increased sharpness and overall increased enjoyment.

[huhn]

@Arm3nian

nnedi3 works best on aliased source like anime or CGI animations not as good on normal film.
so anime give needi3 a lot to work with. on a more blurry source the effect is not as big.
and is not like everything is about sharpness. nnedi3 is awesome because it doesn't create ringing.

[Arm3nian]

Maybe, but I'm not comparing which initially looks better with nnedi3. I'm saying as I raise neurons I see more of a difference in bluray than anime. I don't see any difference in anime going higher than 16.

[fredgido]

I wanted to report that if you activate High Precision Event Timer (HPET) on windows the screen refresh rate gets accurate to to the 4th decimal case.(eg:mine stabilizes around 71.92795hz for exactly 71.928HZ) It before was wrong to the 2nd decimal case.:confused:

When in window mode I had some screen tearing always on the same spot (close to bottom) when not using desktop composition. I noticed that in some resolution the tear line was in different places. So I messed a little with my graphics card output timmings it was fixed.:)

HD5850 & i7 920

[*Touche*]

I get 23.99981 for my 24 Hz TV without HPET. Should it have any influence at all? Isn't it only GPU related.

[cyberbeing]

I get 23.99981 for my 24 Hz TV without HPET. Should it have any influence at all? Isn't it only GPU related.

Forcing HPET will have an influence if your default CPU timer (system clock) has jitter or runs slightly too fast|slow. While refresh rates themselves are GPU based, how close software measurements come to your actual refresh rate depends on CPU timer accuracy.

[ryrynz]

This makes no sense. According to this, there is no reason to use madvr because you don't need it. As long as I can see the video I'm good to go. Let me watch my blurays on a cracked tv screen set to 360p, I don't need anything more as long as I can tell what's mostly going on.

I think the point he's trying to drive home is that there are diminishing returns on scaling that depending on your system may not be worth the trade off required.


waste of resources? So not using my full gpu isn't a waste but using the potential is... No gpu is going to degrade as long as it is being fed a safe voltage and is properly cooled. Electronic components are meant to run 24/7, running 16 vs 32 isn't going to do anything to your gpu life span. You mention the resources being used for something else, but there isn't anything else for blurays, I already have all of the image enhancing settings turned on/configured.

This is your personal preference, if I had the power to run nnedi 256 I might just do it. There are some minor differences I can see over 64 which is my current luma setting for less than 720 content. Chroma is a complete waste of resources in mine and other people's opinions but of someone wishes to run that then it's their power bill..


And as a final note, I can see a little sharpness difference between 16 and 32. Not sure what content or what type of screen you have, but just because you can't see a difference doesn't mean others can't. There is obviously diminishing returns as you go up in neurons, but this exists for everything. If I upgrade from a $500 pair of headphones to a $1000 pair am I going to get 2x better sound quality? Probably not. Doesn't mean there isn't a difference. Just because 32 uses more gpu power for a smaller increase in IQ doesn't mean it's wrong to use it, it's called being an Enthusiast.

I'm very happy with 64 neurons. I doubt I'd notice a difference at 256 during movement but up close on particular frames I can see better line resolving and I'm always interested in higher PQ..


Just to add, I notice from your username that you must watch a lot of anime. Animation by nature is already very sharp. Bluray is shot on film, which is naturally grainy. Sharpness goes a long way for bluray from the movies that I have watched. Everyone's needs and wants are different. Saying 32 neurons is useless compared to 16 is ignorant. You're on a laptop, going from 16 to 32 would just create more heat which will burn your legs or table and increase fan noise and drain your battery. Those two effects are not worth a slight increase in an already sharp piece of content. I'm on the most power efficient htpc I could build that runs cool by default, and I am willing to trade a little power draw for increased sharpness and overall increased enjoyment.

Do remember there is a lot of animation out there at 640x480 and lower which benefits greatly from nnedi. In fact I don't really need to run my Avisynth script for basically doing the same sort of thing any more.

Everyone has their views on what's "worth it" although I would suggest animeviewer looks into some comparisons of 16 vs 64 neurons.. With a 750ti I'd recommend cranking it up.
BTW the latest beta drivers gave me a small boost in rendering times ^_^

Just a tip.. Don't bother with nnedi on 720 content on a 1080 screen. Completely pointless from the tests I did. I use Jinc 3 exclusively for up scaling above 720 and also for further up scaling after nnedi (although lanczos is another great alternative) Of course both with AR enabled.

[Arm3nian]

I think the point he's trying to drive home is that there are diminishing returns on scaling that depending on your system may not be worth the trade off required. This is your personal preference, if I had the power to run nnedi 256 I might just do it. There are some minor differences I can see over 64 which is my current luma setting for less than 720 content. Chroma is a complete waste of resources in mine and other people's opinions but of someone wishes to run that then it's their power bill.. I'm very happy with 64 neurons. I doubt I'd notice a difference at 256 during movement but up close on particular frames I can see better line resolving and I'm always interested in higher PQ.. Do remember there is a lot of animation out there at 640x480 and lower which benefits greatly from nnedi. In fact I don't really need to run my Avisynth script for basically doing the same sort of thing any more. Everyone has their views on what's "worth it" although I would suggest animeviewer looks into some comparisons of 16 vs 64 neurons.. With a 750ti I'd recommend cranking it up.
BTW the latest beta drivers gave me a small boost in rendering times ^_^ Just a tip.. Don't bother with nnedi on 720 content on a 1080 screen. Completely pointless from the tests I did. I use Jinc 3 exclusively for up scaling above 720 and also for further up scaling after nnedi although lanczos is another great alternative.. Of course both with AR enabled.

Exactly my point. Saying running higher than 16 is useless would be valid if there was 0 difference in IQ, but there is. Now if that small difference isn't worth more power draw and heat to you then don't run it. But it is to me since I'm after the best IQ I can get with my system. Plus the 750ti is extremely power efficient.

I watch mostly HD content, and there is obviously diminishing returns trying to improve HD quality over SD with madvr, but as long as I can see a difference, I'll do it.

[TheRyuu]

I noticed that the madvr binaries are not flagged as dynamic base (ASLR) or nx compatible (DEP).

So which is the culprit here, an ancient Visual Studio or mingw (binutils)?

[cyberbeing]

At least in the case of madVR.ax, it appears to be linked with Visual Studio 2005 defaults (/subsystem 4.0, aslr & nx disabled).

The rest of the madVR binaries are Delphi, and a quick google search suggests it also has both options disabled by default.

[TheRyuu]

At least in the case of madVR.ax, it appears to be linked with Visual Studio 2005 defaults (/subsystem 4.0, aslr & nx disabled).

The rest of the madVR binaries are Delphi, and a quick google search suggests it also has both options disabled by default.

Ah, well got it (possibly) half right.

There really isn't any reason not to have them set though. I suppose it matters less for DEP since that only has to be set by the main executable but still, doesn't look good.

Pretty sure binutils still defaults to "disabled" as well which I never understood.

I wanted to report that if you activate High Precision Event Timer (HPET) on windows...

Doesn't Windows automatically chose the best source for the system timer at startup (e.g. TSC constant rate)? Pretty sure it's not a great idea to force that on (presumably with bcdedit).

[cyberbeing]

Doesn't Windows automatically chose the best source for the system timer at startup (e.g. TSC constant rate)?

Windows uses a hybrid timer which it attempts to sync lower accuracy TSC (invariant constant rate or otherwise) to HPET when higher accuracy is needed. This improves the accuracy of TSC, but such a hybrid approach will never achieve the full accuracy and stability of running HPET alone. Does this matter? Not usually, and Microsoft's default hybrid approach has some advantages over forcing HPET, unless of course your CPU's TSC timer is unstable for some reason.

This is easy to test in madVR, since it always requests timers with the highest accuracy available of 0.5ms.
1) Does madVR always report the exact same refresh rate each time you launch a video or reboot, after given a few minutes to settle?
2) Does madVR always report nearly the same 'clock deviation' and 'dropped/repeated frames every' counter, after given a few minutes to settle?

If the answer to both of these are no, that could be a case in favor of forcing HPET, as it will often improve the stability of various clocks and measurements. By the same measure, if you force HPET and see no change in madVR readings, it would be highly recommended to just stick with the Windows defaults. As always, if you intend of forcing HPET long-term, you should re-test any computation intensive applications you may run (i.e. encoding, decoding, avisynth filters, gaming, etc) as well as UI responsiveness, in order to ensure performance remains neutral or improves.

[leeperry]

Microsoft's default hybrid approach has some advantages over forcing HPET, unless of course your CPU's TSC timer is unstable for some reason. .
And what are they exactly? All newest CPU's support Invariant TSC and I was told that its advantage over HPET was a lower latency?

I use an app a friend of mine wrote that permanently forces 0.5ms, I vaguely remember you saying that forcing 0.5 would force HPET anyway?

I don't mean to get OT but I was all for forcing HPET, the thing is that many ppl claim that it's a debugging feature and that Invariant TSC is just as accurate without the extra latency so I'm not sure what is best for A/V purposes anymore :o

http://forums.guru3d.com/showpost.php?p=4876444&postcount=806
7ns=TSC used, because it means shorter access to the hardware and TSC is the shortest one, compared to HPET.
While QueryPerformanceCounter benefited using the HPET/TSC when compared to ACPI PM timer, these days the HPET is outdated by the invariant TSC for many applications.
http://forum-images.hardware.fr/images/perso/1/chewee297.gif

[nevcairiel]

The answer is quite obvious:
If you have no problems, you should just leave it alone.

Additionally, this is not the place to discuss windows timings.

[leeperry]

Yep, mVR and foobar work just fine with invariant TSC on W7SP1 and this timer was one of my main motives to upgrade so be it :)

This said, the guys in that thread claim that W7/W8 react differently:
this might be a bug for the TSC+HPET sync for Windows 7, as for Windows 8+, you will get 15.625 ms(64 tick) compare to Windows 7 which is 15.6ms.

And if you set the Timer Resolution as 1ms for Windows 8+, it will also increase itself as 1.001ms too.
Maybe a good reason to bite the W8 bullet, provided that it would even matter as far as mVR/foobar are concerned.

[Morku]

Hi, I have some 16fps and 18fps footage, typical known for Super 8 or Standard 8 cine films (or very old movies).
My TV supports these modes: 1080p23, 1080p24, 1080i25, 1080p29, 1080p30, 1080p50, 1080p59, 1080p60
So I inserted in "display modes". Smooth motion is on.

When I start a movie with 16/18fps, it always switch into 23.976Hz mode.
For 16fps the specs looks fine.
16fps23Hz: http://s1.directupload.net/images/140821/9bk4hwqc.png

When I start 18fps, the specs says, there is a frame repeat after every 0.85 seconds!
18fps32Hz:
http://s7.directupload.net/images/140821/3782tt9f.png

When my tv stays at 60Hz (I put 1080p18 as 'workaround' in display modes. It doesn't switch for 18fps.) there is not such problem.
18fps60Hz:
http://s14.directupload.net/images/140821/oiu5dai9.png

In my opinion the better choice.
Is that a correct behavior? Or do I something wrong? Am I wrong?

To be complete. Thats the result for 16fps60Hz:
http://s14.directupload.net/images/140821/9pp68lj7.png

[Mano]

Could you clarify this? Which subtitle filter(s) were you using before installing XySubFilter? If it was the MPC-HC ISR, are you sure you've disabled it after installing XySubFilter?

First make sure you've disabled MPC-HC Playback -> 'Use Built-in Subtitle Renderer'. Then load a video, right-click the video window, and check the Filters context menu to ensure that "XySubFilter" is present. The 'Use Built-in Subtitle Renderer' option is the setting for enabling and disabling the MPC-HC ISR. When the ISR is disabled, XySubFilter should be used if it is installed.

When you say this only occurs with madVR, do you mean:

Out-of-memory Crash - both madVR + MPC-HC ISR and madVR + XySubFilter?

Working - both EVR-CP + MPC-HC ISR and EVR-CP + XySubFilter?

If you are unsure, please re-test these four cases. Assuming this really does not occur with EVR-CP no matter what subtitle renderer is used, then your CPU queue setting in madVR must be at fault if you are hitting the 32-bit process limit on system ram.


madVR settings -> rendering -> general settings is where you'll find the queue settings. Resetting madVR to defaults wouldn't be a bad idea, if you haven't already tried this.

Lowering the 'CPU queue' will reduce System RAM usage when using XySubFilter, but increases the chance of dropped frames during slowdowns. Though really, the only reason any extreme amount of ram should be used at all, is if script contained extremely massive objects which extended beyond the screen size. Scripts which even come close to the 32-bit process limit at 2560x1440 with madVR and XySubFilter defaults are extremely rare, so I'm a bit surprised you are having issues.

I'd really need a sample of the script which is causing your problems, along with the resolution of your monitor. Bad come to worse, if you enable the "Render to Original Video Size" option in XySubFilter, you should never hit the 32-bit process limit on any released script (i.e. no one sane would release a script which would cause out-of-memory errors at original size). But if you are saying everything works with EVR-CP, then just lowering madVR's queues should be enough.

I remember installed it long before when i install madvr and xySubFilter is in my start menu list.

I just tried uninstall everything then update mpc, madvr and xySubFilter to the lastest and it no longer crash. So either it was bug with old version or i changed some setting wrongly. Too bad i no longer keep the old install file so i cannot check what exactly cause the crash.

the script is too large to put on pastebin. If you want it, i uploaded it to mediafire. The place where i crash is 19:00

https://www.mediafire.com/?4p8f408zwntrgfd

[vivan]

Kaylith ZnT? I'm not suprised :)

What is you monitor resolution? 1080p or higher?

[cyberbeing]

So either it was bug with old version or i changed some setting wrongly. Too bad i no longer keep the old install file so i cannot check what exactly cause the crash.
Thanks for the sample. The way that script used /iclip is rather inefficient in terms of memory usage and performance with VSFilter-based filters, compared to using /clip to do the same thing. So while it may drop frames in some parts if scaled to 2560x1440, it definitely doesn't look like that should cause out-of-memory crashes, even with larger than default queues and cache sizes with the latest XySubFilter build (as you seem to have found). Probably not a bug, since starting with Beta2 there were a couple changes which reduced memory usage compared to Beta1, so I'd assume it was just that.

[Kevin Gilmore]

Hi, does anybody know why I cannot convince Maxwell 860M to HVA decode? >

I tried the 'forceVendor' 'Nvidia' which was straightforward, and renaming the exe MPC-HC file.
While that 'alerted' or 'woke up' the gpu up and ram is used, the VPU section, GCardHW, etc is not triggered. It seems I've crossed a threshold, but the Maxwell gpu is playing stupidly hard to get. [NOTE: THIS COULD BE VERY PROBLEMATIC FOR FUTURE YOUTUBE VIDEOS / SUPPORT e.g. I did NOT purchase an 860M to perpetually watch/decode 4k on my yeasty hd 4600]

@Nekromantic What registry strings were entered to enable the Maxwell mobile chips/what are the correct registry stings?
How on earth WERE you Triumphant? 'custom registry hack? Am I to understand it's only a nominal triumph as madVR Chroma upscaling + Chroma image doubling is MAxwell Hardware video accelerated - pertaining to madVR, as well?

Thanks to all/any who know of a fix; I bet there are at least a few others out there who've this question/ harbor similar disconcertions.

I just got a new laptop running Windows 8.1 and on latest Nvidia GTX860M drivers.
I set up LAV video to use software render and installed MadVR. When running on Intel card MPC HC gets very high 150ms render time with and without hardware rendering.
This is when using NNED13 for chroma upscale.
I can use Jinc 3 taps fine.

I should be able to use NND13 on a i7 laptop with 860m.

[vivan]

the VPU section, etc is not triggered.It has nothing to do with madVR, it's the decoder.

If you want to use H/W decoder - then you should select DXVA or nVidia CUVID decoders in Lav Video Decoder settings. If you have problems with them - welcome to http://forum.doom9.org/showthread.php?t=156191

[Mano]

Kaylith ZnT? I'm not suprised :)

What is you monitor resolution? 1080p or higher?

yep 2560x1440 27" dell

[Anime Viewer]

Hi, does anybody know why I cannot get Optimus for/on Maxwell 860M to run?

I tried the 'forceVendor' 'Nvidia' which was straightforward, and renaming the exe MPC-HC file.

How can one What registry strings were entered to enable the Maxwell mobile chips/what are the correct registry stings?


As vivian said madshi doesn't believe it is a madVR issue.

Other 860m Optimus users have stated that they've only got video to playback with the Nvidia gpu forced in MPC-HC if they went into MPC Options -> Output and set DirectShow Video to "system default" and the RealMedia Video and QuickTime video to Video Mixing Renderer 7. That both effective disables madVR and DirectX9 features. madshi has recommended people reinstall the DirectX 9 runtime files, but last I heard no one has had success getting the runtime to reinstall.

As far as madVR adding the registry key for the 860m: you'll need to have madVR set as the DirectShow Video output in MPC Options, and have one of the NNEDI3 options selected. That will get your key added, but it will not do anything to get the 860m to run properly.

You could make a separate thread elsewhere in the software players forum to see if any 860m users have found a workaround, but unfortunately right now a solution may be necessary from Microsoft and/or Nvidia's end. Given Nvidia's history when it comes to fixing madVR/MPC/non-game related bugs there may never be a fix. :(

Some people believe it is related to the following:

The following code snippet caused the error:

VMR9NormalizedRect r1 = new VMR9NormalizedRect(0, 0, 0.5f, 1);
VMR9NormalizedRect r2 = new VMR9NormalizedRect(0.5f, 0, 1, 1);
hr = (HRESULT)mix.SetOutputRect(0, ref r1);
hr = (HRESULT)mix.SetOutputRect(1, ref r2);

If a VMR9NormalizedRect gets initialized with any parameters apart from 0, 0, 1, 1 it will only display a black screen.


I've also read that on ZoomPlayer people were able to get around the VRM9 issue by selecting "Enable YUV mixing on all VMR Video Renderers (Less CPU overhead on some configurations)". I'm not sure if there is a similar setting in MPC/LAV/madVR that might do the same. (Edit: looks like YUV Mixing may already be checked by default in the directshow video VMR settings area in MPC, so that may not be a fix in this case).

[StinDaWg]

Just a tip.. Don't bother with nnedi on 720 content on a 1080 screen. Completely pointless from the tests I did. I use Jinc 3 exclusively for up scaling above 720 and also for further up scaling after nnedi (although lanczos is another great alternative) Of course both with AR enabled.
I couldn't disagree more. For 720p->1080p scaling, NNEDI3 is the only upscaler that does not produce jagged lines and aliasing which is especially noticeable with text. The overall picture is also sharper than Jinc with less artifacts. This is on a 60" tv. YMMV.

I absolutely love the way NNEDI3 makes 720p content look. On a good source the difference between the 720p encode and 1080p source can be difficult to distinguish at times.

[ryrynz]

I couldn't disagree more. For 720p->1080p scaling, NNEDI3 is the only upscaler that does not produce jagged lines and aliasing which is especially noticeable with text. The overall picture is also sharper than Jinc with less artifacts. This is on a 60" tv. YMMV.


Jagged lines? Ain't no jagged lines with Jinc man, it's like the nicest upscaler to edges there is.

NNEDI gives great results on text? Yeah I'd agree with you there, so if you want to look at nice sharp NNEDI edges on text then yeah go crank it up, what a serious waste of resources though.

I still stand behind what I said, it does almost nothing on actual footage. It would seem you just lose most of the benefits of NNEDI by downscaling the image down to 1080. (I tested on 1920x1200 as well) There may be the odd scene where NNEDI adds something worthwhile to 720 content but I have yet to find a scene where it does.


I absolutely love the way NNEDI3 makes 720p content look.

Feel free to post a screenshot that proves it's worth enabling. I generally don't watch the credits roll enough to warrant enabling NNEDI on anything 720 or higher not to mention how taxing it is on the GPU for such an insignificant overall PQ improvement.

[Neet009]

I found something that are concered with madTPG and ArgyllCMS. When using dispcal.exe and madTPG to calibration the monitor,the madVR settings "how many video frames shall be presented in advace"could not set more than 10, otherwise it will show the error message with dispcal.exe and finish the work. Windowed mode and FSE mode have the same issue. It works well when setting <= 10.

Is it a bug?

OS:win7 x64
madVR:v0.87.10
ArgyllCMS:1.6.3
graphic card:750Ti
meter:colormunki design

[StinDaWg]

Jagged lines? Ain't no jagged lines with Jinc man, it's like the nicest upscaler to edges there is.
Nope.

Jinc3 vs. NNEDI3
http://abload.de/img/j3ar-dial7ceum.png http://abload.de/img/nnedi3-dialbvdjm.png

Edit: 2nd images too big as this forum doesn't auto re-size to width.
http://abload.de/img/mn-full55c1a.jpg http://abload.de/img/nnedi3-fullvdfok.jpg

[fairchild]

So if you watch alot of gaming footage with alot of HUD's and text then yes, there is a clear advantage to NNEDI3 even for 720p-1080p. But thankfully I don't do that and stick to movies and tv shows. (guess it could be good for watching ESPN but also don't do that either, I just watch it live through my cable box so no need to upscale it with MadVR and my HTPC.)

[Niyawa]

I'd like to ask a quick question (which my search results didn't get me anything).

When using MPC-HC, by checking the "Statistics" while playing a video with madVR, the specified framerate for 23.976 will be 60 (http://i.imgur.com/SU8aW3j.jpg), yet with EVR-CP it will be 23.976 (http://i.imgur.com/fd4k0zh.jpg) instead. It seems that madVR likes to render frames that would otherwise just be repeated instead. Is there a reasoning behind this?

[Mangix]

I'd like to ask a quick question (which my search results didn't get me anything).

When using MPC-HC, by checking the "Statistics" while playing a video with madVR, the specified framerate for 23.976 will be 60 (http://i.imgur.com/SU8aW3j.jpg), yet with EVR-CP it will be 23.976 (http://i.imgur.com/fd4k0zh.jpg) instead. It seems that madVR likes to render frames that would otherwise just be repeated instead. Is there a reasoning behind this?

IIRC, it has to do with displaying frames at a certain time. That is: you can send 24 frames to the GPU per second and let it decide when to double or madVR can do it instead and send the GPU 60fps. Probably necessary for SmoothMotion.

the old behavior can be enabled with the old path(disabling "present several frames in advance"). Be warned that some features can/will break if you do this.

[Anime Viewer]

I'd like to ask a quick question (which my search results didn't get me anything).

When using MPC-HC, by checking the "Statistics" while playing a video with madVR, the specified framerate for 23.976 will be 60 (http://i.imgur.com/SU8aW3j.jpg), yet with EVR-CP it will be 23.976 (http://i.imgur.com/fd4k0zh.jpg) instead. It seems that madVR likes to render frames that would otherwise just be repeated instead. Is there a reasoning behind this?

Does the screen you're watching on support 23hz (or 24hz) as a refresh rate, and do you have that entered in madVR display modes in the list all display modes madVR may switch to: ?

I have statistics displaying on a tv that supports both frame rates, it has both in the list, and it is showing a frame reading of 23.90 and up (it slowly ticks up as the 23.976 video plays, so it would probably get to 23.97 if I leave it running).

[cyberbeing]

Be warned that some features can/will break if you do this.
This is false. No features can/will 'break' if you disable Present in Advance.

In any case, when Present in Advance is enabled madVR presents a frame every vsync interval. If disabled, madVR presents a frame every movie framerate interval (if smooth motion disabled). This is why MPC-HC statistics may show a discrepancy if you have some of these settings enabled and use higher refresh rates than the video.

[sneaker_ger]

This is false. No features can/will 'break' if you disable Present in Advance.
Try native DXVA in FSE mode without it.

[huhn]

Try native DXVA in FSE mode without it.
backbuffer is not filling up but works fine.

[cyberbeing]

backbuffer is not filling up but works fine.

Backbuffer fills just fine here with DXVA Native and disabled Present in Advance.

What kind of problems are you referring to sneaker_ger?

Edit: Ah you are talking about this bug I reported (http://bugs.madshi.net/view.php?id=211) which occurs when you disable FSE Present in Advance. If you only disable Present in Advance in Windowed mode and/or disable automatic FSE entirely, you won't trigger that bug.

Edit2: For example, if you Disable Automatic FSE and FSE Present in Advance, open a video using DXVA2 native, and then re-enable Automatic FSE via options or keyboard hotkey during playback, you won't trigger this bug and will have FSE (old mode) with DXVA native. This setting bug notwithstanding, I maintain that disabling the Present in Advance function itself does not break any madVR features.

[ryrynz]

Nope.


Jinc isn't creating jagged edges.. you're starting off with jagged edges. Jinc isn't a smart neural resizer so of course it isn't going to recreate beautiful edges like NNEDI does, but Jinc is by no means a bad resizer on edges.. it's quite soft in comparison to many others..

[Mangix]

This is false. No features can/will 'break' if you disable Present in Advance.

try in exclusive mode.

[cyberbeing]

try in exclusive mode.

Works fine with FSE as long as you bypass the bug I mentioned above in my edits.

[Niyawa]

IIRC, it has to do with displaying frames at a certain time. That is: you can send 24 frames to the GPU per second and let it decide when to double or madVR can do it instead and send the GPU 60fps. Probably necessary for SmoothMotion.

the old behavior can be enabled with the old path(disabling "present several frames in advance"). Be warned that some features can/will break if you do this.
I see, thanks for the answer.

Does the screen you're watching on support 23hz (or 24hz) as a refresh rate, and do you have that entered in madVR display modes in the list all display modes madVR may switch to: ?

I have statistics displaying on a tv that supports both frame rates, it has both in the list, and it is showing a frame reading of 23.90 and up (it slowly ticks up as the 23.976 video plays, so it would probably get to 23.97 if I leave it running).
Nope, my screen is a laptop one that only supports 60 Hz.

[StinDaWg]

Jinc isn't creating jagged edges.. you're starting off with jagged edges. Jinc isn't a smart neural resizer so of course it isn't going to recreate beautiful edges like NNEDI does, but Jinc is by no means a bad resizer on edges.. it's quite soft in comparison to many others..
Most 720p content doesn't have jagged edges, but can appear jagged after upscaling to 1080p. Jinc is not bad but it's not as good as NNEDI3 when it comes to angled lines, that's not up for debate.

So if you watch alot of gaming footage with alot of HUD's and text then yes, there is a clear advantage to NNEDI3 even for 720p-1080p. But thankfully I don't do that and stick to movies and tv shows. (guess it could be good for watching ESPN but also don't do that either, I just watch it live through my cable box so no need to upscale it with MadVR and my HTPC.)
I don't have any screenshot comparisons but I'm just going by what I see. Even filmed content like Homeland looks better with NNEDI3, it's sharper with more fine details, and there is less artifacts in the picture. NNEDI3 barely taxes my 7850 so there's no reason not to use it.

[*Touche*]

Most 720p content doesn't have jagged edges, but can appear jagged after upscaling to 1080p. Jinc is not bad but it's not as good as NNEDI3 when it comes to angled lines, that's not up for debate.


I don't have any screenshot comparisons but I'm just going by what I see. Even filmed content like Homeland looks better with NNEDI3, it's sharper with more fine details, and there is less artifacts in the picture. NNEDI3 barely taxes my 7850 so there's no reason not to use it.

Why don't you take some screenshots where you see the difference?

[cvrkuth]

My source video is 1080p mkv, BDRemux m2ts etc. My TV is also 1080p. Does the scaling algorithm matter?

[huhn]

My source video is 1080p mkv, BDRemux m2ts etc. My TV is also 1080p. Does the scaling algorithm matter?

chroma yes.
image no.